Meletios Niros If the results of the EFA divided the construct into 2 sub-factors, then you should just report the EFA results, then run CFA with 2nd-order.

Are your measures reflective or formative? If they are reflective, you could use first-order, if they are formative you should use second order.

Yes that was what we did. But the model fit was found below threshold when using the uni scale of the 2nd order within the model. Then we tried the seperate dimensions of the 2nd order and the statistics were excellent. Is this correct? Or should we run the model with the 3nd order scale anyways (even if the model fit is not good)?

I suspect that the measurement variables have a reflective character. Otherwise, Amos would be difficult to apply. I would forgo EFA for an established scale and analyze the scale with ESEM/ECFA or other SEM techniques. You obviously already have a good result that contradicts the literature. The rest of the question is not entirely clear to me. Whether you include first order or second order constructs in SEM is a matter of research design, isn't it? Best. Christian

Long She thank you very much for your comment. This is something that is clearly described in SEM LR.

Olivier Furrer the intercorrelations of Brand Experience dimensions are mixed. This is a problem. So I would say formative, thus second order looks better. The point is the low model fit. Is it justifiable to demonstrate figures below threshold???

Christian Arnold the intercorrelations of Brand Experience dimensions are mixed. This is a problem. So is it formative or reflective? Using second order I get poor model fit. EFA for an established scale and analyze the scale with ESEM/ECFA or other SEM techniques seems to be a viable option. I'll check and I will let you know. Many thanks!!!

I can't see any connection there. Formative roughly refers to measuring the causes of the phenomenon and reflective refers to measuring consequences.

Christian Arnold miele danke!

You should not let your data drive your conceptualization of a construct, it should be the other way round. Please refer to Brakus (2009) for the conceptual definition of brand experience and see if your research uses a similar conceptualization to interpret your results with reference to Brakus (2009).

As for the choice between first vs. second order, it would depend on how you wish to treat brand experience in your conceptual model - as an overall measure of brand experience or a combination of its individual dimensions - to study their impact on other variables in your model.

Hope this helps!

Piyush Sharma thank you for your point of view. It seems promising!

You should first consider whether your 1st order constructs are reflective/formative. Then consider whether your potential 2nd order construct is reflective/formative and how this makes sense theoretically. High intercorrelations between your 1st order constructs could indicate the need for a 2nd order reflective construct (Brown, 2015, p.40). Also, in your EFA for the 1st order constructs check for cross-loadings, they could indicate you need to group factors with cross-loaded items in a reflective 2nd order construct. Sometimes, if you 'merge' the dimensions of a 2nd order construct into a single 1st order construct, you might get a lower fit for the merged construct and for the overall model.

You should also compare the model fit of your 1st order CFA (where you rule out 2nd order constructs) with the model fit of a 2nd order CFA (where you include your 2nd order construct) (Wright et al., 2012).

For choosing between reflective/formative modeling and for comparing the advantages of 2nd order modelling I recommend reading more from:

Koufteros, X., Babbar, S. şi Kaighobadi, M., 2009. A paradigm for examining second-order factor models employing structural equation modeling. International Journal of Production Economics, 120(2), pp.633-652.

Bagozzi, R.P. şi Yi, Y., 2012. Specification, evaluation, and interpretation of structural equation models. Journal of the Academy of Marketing Science, 40(1), pp.8-34.

Wetzels, M., Odekerken-Schröder, G. şi Van Oppen, C., 2009. Using PLS path modeling for assessing hierarchical construct models: Guidelines and empirical illustration. MIS Quarterly, 33(1), pp.177-195.

Wong, C.S., Law, K.S. şi Huang, G.H., 2008. On the importance of conducting construct-level analysis for multidimensional constructs in theory development and testing. Journal of Management, 34(4), pp.744-764.

Wright, R.T., Campbell, D.E., Thatcher, J.B. şi & Roberts, N., 2012. Operationalizing multidimensional constructs in structural equation modeling: Recommendations for IS research. Communications of the Association for Information Systems, 30(1), pp.367-412.

Brown, T. A. (2015). Confirmatory factor analysis for applied research. Guilford publications.